Documentation update tool_calling: mapping back to function from response #20373
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @cronoik-inceptionai, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request focuses on improving the clarity and best practices within the tool-calling documentation. It refines an existing code example to showcase a more flexible and maintainable pattern for dispatching functions based on model responses, moving away from hardcoded function calls to a dynamic lookup mechanism.
Highlights
- Documentation Update: The example code for tool calling in
docs/features/tool_calling.mdhas been updated. Previously, the example directly called a specific function (get_weather) after parsing the tool call. The updated example now demonstrates a more robust and scalable approach by using a dictionary (tool_functions) to map the tool call's name to the corresponding function, then invoking it dynamically.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
mergify
bot
added
documentation
There was a problem hiding this comment.
Code Review
This pull request updates a documentation example for tool calling to use a more robust dynamic function dispatch pattern. The change is a clear improvement. My review includes a suggestion to further enhance the example by adding error handling for cases where the model might return an unexpected function name, which would make the example more resilient and educational for users.
| print(f"Function called: {tool_call.name}") | ||
| print(f"Arguments: {tool_call.arguments}") | ||
| print(f"Result: {get_weather(**json.loads(tool_call.arguments))}") | ||
| print(f"Result: {tool_functions[tool_call.name](**json.loads(tool_call.arguments))}") |
There was a problem hiding this comment.
This change is a great improvement to the documentation as it demonstrates a more robust and scalable pattern for tool calling.
However, the current implementation tool_functions[tool_call.name] is susceptible to a KeyError if the model hallucinates a function name that is not defined in the tool_functions dictionary. While this might be acceptable for a simple example, making it more robust would be highly beneficial for users who might adapt this code for production use.
Consider adding error handling to gracefully manage cases where the function name is not found. Here's a more robust approach:
# Safely get the function from the dictionary
func_to_call = tool_functions.get(tool_call.name)
if func_to_call:
# It's also good practice to handle potential JSON and argument errors
try:
arguments = json.loads(tool_call.arguments)
result = func_to_call(**arguments)
print(f"Result: {result}")
except (json.JSONDecodeError, TypeError) as e:
print(f"Error calling function '{tool_call.name}': {e}")
else:
print(f"Error: Function '{tool_call.name}' not found.")Since this change would expand a single line into multiple, I'm providing it here for your consideration rather than as a direct suggestion. This would help users writing production code to be aware of potential pitfalls.
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
…onse (vllm-project#20373) Signed-off-by: Jinzhen Lin <[email protected]>
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
The example code shown in the documentation was calling the function directly after the response, but it is an better example to use the tool_call dictionary to fetch the function to execute.
Test Plan
Not required.
Test Result
(Optional) Documentation Update